Data Clustering Methods for Recognition of Endoscopic Images in the Problems of Computer Medical Diagnosis

This paper presents the results of the analysis of existing methods for clustering data obtained during endoscopy of a larynx. A modification of the Viola-Jones method for image recognition using the flexible exit criterion is proposed. The Viola-Jones method explores all areas in the image and decides whether the recognized area belongs to the desired one by passing through a classified cascade. Endoscopic images have a large number of features, such as flare, noise, etc., which degrade the quality of recognition. To improve the quality of recognition, clustering with a flexible exit criterion was proposed, which satisfies the scalability criteria: changing the decision of the solution, instead of moving to another recognition area. It has been established that the proposed modification of the Viola-Jones method shows higher recognition results for endoscopic images.